[Keynote Talk] AUTOMATIC DETECTION OF GENERATED VOICES AND FACES – ASVSPOOF AND DEEPFAKE DETECTION

講演者：Junichi Yamagishi
会議名：The 23rd International Conference on Speech and Computer SPECOM-2021
主催者：The International Conference on Speech and Computer (SPECOM)
開催地：オンライン
URL：http://specom.nw.ru/2021/keynotes/

In this talk, we introduce our recent projects on automatic detection of generated voices or faces. For detection of computer-generated voices, we have constructed a large dataset that contains various types of synthesized and converted speech and organized the ASVspoof challenge where over 50 teams competed for the best anti-spoofing performance. Based on analysis of the challenge results, we show important lessons learned from both the attacker and defender perspectives. In the latter part of the talk, we describe another project on deepfake detection and show new countermeasure models using capsule networks or multi-task models that indicate manipulated facial regions for improving explainability.